Discussing the ambiguous article and showing the impact of the maximum recovery period
During intensified regulatory oversight, ECB found that banks could keep defaults with high losses open forever which kept many high losses out of the modelling dataset, leading to seemingly low losses in the portfolio. To prevent this, the latest AIRB models ought to be developed including a maximum recovery period: a certain period after which the bank is not allowed to estimate any recoveries for the purpose of modeling. Having these cases open for a long time now leads to having high loss cases in the dataset.
In our opinion, this is a debatable piece of regulation, especially for collateralized asset classes like residential mortgages. The article in the regulation can be interpreted in at least two ways and the additional Q&A text from EBA has not provided a definite answer on the implementation. Without question, the impact on LGD is very large.
First we will introduce the maximum recovery period and discuss the options for interpretation. Then, we assess the impact and finally address some argument for the discussion of interpretation, including a view from the regulator.
For the modeling process of AIRB LGD models, banks should take into account all defaulted exposures in the dataset . A part of the data will consist of unresolved cases, for which realized loss is currently unknown. Sometimes open default cases have been in default for a very long time and it is questioned whether there will be any recoveries at all. To address this problem and avoid cases are kept open on purpose, the EBA introduced the maximum recovery period (MRP) for banks. The MRP is defined as:
´A period that reflects the expected period of time observed on the closed recovery processes during which the institution realises the vast majority of the recoveries´.
For the purpose of modeling, observed losses for unresolved recovery processes should be adjusted for the maximum recovery period. How losses for open defaults should be adjusted is split into two types of open cases based on the current time-in-default:
Type 1) For cases that have passed the MRP at the moment of modelling no recoveries should be estimated:
Type 2) For cases that have not yet passed the MRP at the moment of modeling the rules below should be taken into account for constructing an observed loss:
Two interpretations of the article on the MRP
Article 158 of the guidelines in which the MRP is explained states that:
a) they should take into account all observed costs and recoveries;
b) they may estimate future costs and recoveries, both those stemming from the realisation of the existing collaterals and those to be realised without the use of collaterals within the maximum period of the recovery processes.
We think that the interpretation of the second statement (art. 158b ) is open for some debate. Let us discuss two different interpretations below on how to construct observed losses for unresolved cases.
They may estimate future costs and recoveries, both
those stemming from the realisation of the existing collaterals
those to be realised without the use of collaterals
within the maximum period of the recovery processes.
Here both recoveries from collateral and other recoveries (so all recoveries) should only be estimated up to the MRP (conservative interpretation)
those stemming from the realisation of the existing collaterals
those to be realised without the use of collaterals within the maximum period of the recovery processes.
Here it is allowed to make an estimation for recoveries from the collateral, even after the MRP (progressive interpretation)
Reading both statements, which one do you thinks is more plausible? We think that the right, progressive, interpretation is more sensible for this piece of regulation for various reasons, which will be discussed in a later paragraph.
First, we want to look at the large impact the interpretation has on the LGD of an average Dutch mortgage portfolio.
The difference in LGD predictions stems from the (LRA) calibration where unresolved cases are included. Depending on the length of the modeling dataset and the moment of modeling, up to 20% of the data can have an unresolved status. If the moment of modeling is during a crisis with many defaults, this percentage could be even higher.
Part of these open cases will be of type 1, as previously defined. These cases should write off all open exposures, i.e. no estimation of future recoveries is allowed. Typically, the vast majority of a mortgage loan is recovered by selling the collateral at the end of the default. Therefore, the fact that the case is still open must mean that the collateral is not yet sold. And as the collateral is not sold yet, the open default will have an observed LGD of approximately 100%. This is visualized in the figure below.
The relative size of type 1 cases will depend on the MRP. A long MRP will lead to fewer type 1 observations and more type 2 observations. Type 1 observations will have the same observed loss under both interpretations, but due to their high (constructed) loss will have a large impact on the observed LGD. The unresolved type 1 cases and their observed costs and recoveries are illustrated in the figure below.
The remainder are type 2, where the time-in-default has not yet passed the MRP. Looking at the figure below, the conservative interpretation leads only to an estimation for the orange recoveries. The progressive interpretation also allows estimation of the green box, representing recoveries from collateral.
As the chance of exceeding the MRP increase as the time-in-default increases, we observe in the figure below that the difference in observed LGD diverges for the different interpretations. This figure shows the average observed LGD, where the LGD for open cases is constructed without MRP (blue), with the conservative interpretation (orange) and with the progressive interpretation (green). By taking the average of these LGD values per time-in-default months, we can clearly see a time-in-default relation introduced by the MRP, which was not clearly present before.
Irrespective of the interpretation, conforming to the MRP rule drastically increases the observed LGD and thus RWA, for a mortgage portfolio. The progressive interpretation will lead to a lot smaller increase, displayed in the table below.
Why we prefer the progressive interpretation
In our view, it is not likely that a collateral becomes worthless after a certain artificial time period. Is it really sensible, for a mortgage, to deem the collateral worthless after the MRP? If there exists a house that can be sold by the bank, it feels overly conservative to assume that no recovery from collateral is made.
The Bank for International Settlements (BIS) stresses to have a small gap between the regulatory EL amount and the provisions required under Expected Credit Loss accounting approaches. However, as the regulatory EL includes the MRP it will be significantly higher than the accounting ECL. The progressive interpretation will lead to a smaller gap.
Concerning choosing an MRP length, the EBA states the following:
‘CAs should ensure that the concept of the maximum length of the recovery process is not misused and that the lengths specified by institutions are not excessively long for a specific type of exposures.’
It is likely that banks will choose an extremely long MRP in order to reduce the impact of the MRP on RWA. EBA recognizes this and explicitly mentions that misuse should be avoided. The progressive interpretation will lead to similar impact with a more reasonable choice for the MRP.
A regulators point of view
Although we believe that the progressive interpretation of Article 158 is more sensible from an economic and modelling point of view, we observe that the current market practice is to follow the conservative interpretation. Indeed, as pointed out above, to limit capital impact banks are incentivized to choose a long MRP. Our current understanding is that the regulator’s perspective is also to follow the conservative interpretation, overlooking the consequences such as models that are hard to align with risk management intuition and regulatory issues such as ELBE and stage 3 provisions becoming non-comparable.
One way or another, including the MRP will have a large impact on the LGD of a mortgage model, which is overly conservative in our opinion. The interpretation of the regulation also comes with a large impact. Effectively writing off fully collateralized loans in the model dataset makes our modeling minds nervous. We think that banks are in need of an explicit guideline on this topic from the regulator and that the regulator needs to rethink the treatment of unresolved cases and MRP specifically for collateralized loans such as mortgages.
If you want to join our RiskQuest team, please check our current job openings here
 Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures– article 149
 Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures– article 158(b)
 Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures– article 156
 Specifically referring to the long run average calibration for LGD
 Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures– article 157